Christopher Ringhofer

Hardware-Aware Neural Architecture Search for Embedded Audio Effects Simulation


Christopher Ringhofer, M.Sc., has been working as a research assistant and PhD student at the Embedded Systems department of the University Duisburg-Essen since April 2020. His current work focuses on automated search of efficient neural network architectures for signal processing on embedded devices. He is developing a system that uses hardware-aware neural architecture search to construct latency-optimized networks for signal processing of audio data on embedded hardware. His current use case is digital audio effects, e.g. for studio and live applications.

As a research assistant he previously worked on two projects funded by the German Federal Ministry of Education and Research: “KI-Sprung: LUTNet” and “KI-LiveS”.

He received his Bachelor’s Degree for Applied Computer Science with a focus on engineering informatics in 2017 and his Master’s degree in 2020 at the University Duisburg-Essen. During this time he also worked three years as a software engineer in a company focussing on Smart Home products and Internet of Things technologies.

Description of the Talk:

Black-box modelling of non-linear audio effects such as distortion has attracted increasing research interest since the advent of generative audio models such as Google DeepMind’s WaveNet. Despite the advantages of such approaches, the models presented are often too large to be used on resource-limited devices or have impractical inference latency. Therefore, he presents an approach to overcome this problem by automatically finding suitable deep learning architectures for a given audio effect and co-optimising hardware-sensitive costs such as latency.